I’ve recently moved most of my development work to Haskell, and in the main, it’s been extremely successful. High-concurrency work has been a breeze: the only difficulties I’ve hit have been to do with operating system limits, not Haskell’s.

Ruby still has an amazing set of developer niceties, however, and I miss them in Haskell. So, here’s how to get some of the love back, with Guard.

If you haven’t seen it, Guard is a tool that watches your code and tests directories for file changes. When a test changes, the test is run again (and only that test); when code changes, all the tests for that piece of code are run. Being Ruby, it has to rely on conventions to determine that relationship: in general, it expects a one-to-one mapping from spec/foo_spec.rb to lib/foo.rb, and covers its arse by running the whole test suite whenever the set of tests it’s been instructed to run come up green.

Now, we could do better in Haskell, because we statically know all of our code dependencies. However, the build process is a bit involved and I’d need to dig into hspec a little - for the moment I’m just going to run “cabal build && cabal test” every time I detect a change in ./test. Brute force works first time, every time. We will add one little tweak in: I’ve set a flag in my cabal file

   Flag onlyTests
        Default: False

and

   if flag(onlyTests)
     Buildable: False

in the stanza for my executable, as linking it takes a fair while.

Then, when I run

cabal configure -fonlyTests --enable-tests --disable-shared --disable-executable-profiling --disable-library-profiling

so that everything I don’t absolutely need to run the tests is turned off,

cabal build && cabal test

takes 7 seconds from a standing start, and 5 the second time (where the test executable is already built). This could be improved, but I’ll cope for now.

Finally, I install guard (I assume you have a ruby interpreter on your system)

gem install guard guard-shell

and throw in a Guardfile at the base of my Cabal project.

guard :shell do
  watch(%r{.*\.cabal$}) do
    `cabal build && cabal test`
  end

  watch(%r{src/.*hs}) do
    `cabal build && cabal test`
  end

  watch(%r{test/.*hs}) do
    `cabal build && cabal test`
  end

end

Now, when i run “guard” in my haskell project, my tests run every time I save.

This could be improved in many ways - we could run just the tests that guard tells us have actually changed, we could have a final step where we try to build the executable as well (or at least check that it compiles, since the slow step is linking). This can also collide in an unfortunate way with ghc-mod - ghc-mod will periodically save files to check if they compile. If anyone’s looking for a fun little project that would be of active benefit to the community, here’s a place to start.

EDIT

this is a little faster, and will also not fire on flymake temporary files. For that to work well with ghc-mod, you’ll need the latest ghc-mod off hackage - it looks like the version in elpa is pretty old.

 guard :shell do

   def build_command(files)
       "ghc -fno-code #{files.join(' ')} -isrc -e 'return 0' && cabal build && cabal test"
   end
   watch(%r{.*\.cabal$}) do
       `cabal build && cabal test`
   end

   watch(%r{src/.*(?<!flymake)\.hs$}) do |files|
       `#{build_command(files)}`
   end

   watch(%r{test/.*(?<!flymake)\.hs$}) do |files|
       `#{build_command(files)}`
   end

   # because we know exactly what's been changed, we can be a bit
   # trickier about how much to load. If this fails, can always go
   # back to the build_command method
   watch(%r{test/.*(?<!flymake)\.hs$}) do |files|
     cmd = files.collect do |f|
             "ghc -isrc -itest -e 'hspec spec' #{f} test/dummy.hs"
           end.join(' && ')
     `#{cmd}`
   end


 end

and for extra credit, link with the gold linker



Published

03 May 2013

Tags